Ensemble Approach for Capacitance Prediction of Heteroatom Doped Carbon Based Electrode Materials

Abstract

An ensemble approach-based machine learning modeling is used in the current study for unveiling the effect of various electrode parameters on the electrochemical performance of hetero-atom doped nanocarbons. This is achieved using three meta-classifiers in combination with traditional Multi-Layer Perceptron and Random Forest models. The three meta-classifiers used are namely (i) bagging, (ii) classification via regression (CVR) and (iii) multi class classifier (MCC). Amongst these three models, bagging and classification via regression provided greater accuracy in terms of correctly classified instances (%) and area under region of convergence values. The designed models are used to predict class of specific capacitance values. 94.5 % of the considered dataset is classified correctly proving a better accuracy of the designed models. Lowest root mean square value of 0.1787 was obtained for RF model. Compared to the models defined in the literature, the suggested models in this work provide best fit of the experiment and predicted values with highest accuracy and lowest error performance values. The lowest error value for RF and MLP models are 0.18 and 0.19 respectively.